Ziyu Li

COTS target detection task: traditional method

The training data set manually tag (starfish) is positive samples and negative samples (starfish), unified positive samples and negative sample size (64 * 64). Create "train_pos_imgs" and "train_neg_imgs" folders in the root directory to store all positive and negative samples, respectively. After get all the positive and negative samples of preservation, HOG method is used to extract HOG feature, HOG features of extracted into the SVM training in order to obtain the initial classifier.

Long-tailed dataset multi-classification: deep learning method

The project to solve the problem is for dolphins and whales are classified, the original data set is mixed and disorderly, background noise is bigger (such as sea water). Firstly, collate and clean the dataset by EDA, preprocess the dataset images, split the dataset. Due to the serious problem of class sample imbalance in the dataset, the remix algorithm is used to balance the dataset, and then the data processed by remix is standardized and brought into resnet50 (pre-training weight) for training. During training, considering to further reduce the influence caused by the imbalance of class samples, Focal loss is used as the loss function for training, and three-fold cross validation is performed to obtain the final result.

Personal research experience - under the guidance of Dr. Wei Li, Chinese Academy of Sciences

Introduced a meta-learning-based clinical risk prediction algorithm for clinical risk prediction with limited patient electronic health records. The related literature is categorized and compared based on research objectives, methods and experimental analysis. Additionally, potential research opportunities for future work in this area are discussed.

LinFen Vocational And Technical College : Teaching assistant

Assisted in teaching the course of computer principles, and managed the class as the head teacher.